Anthropic’s August 13 update says Claude Tag began using wider channel context, memory and standing instructions to decide when to contribute. Previously, a lightweight classifier considered individual messages in isolation. The revised behavior can reply briefly, begin deeper work, route an update into an existing task or stay silent, while remaining within configured access and permissions.
Anthropic reported roughly a 30 percent improvement in deciding when to respond, but that is a company evaluation, not an independently reproduced score. The announcement also said the added context did not count toward usage or spending limits. Those statements belong to the dated release and should not be treated as a guarantee of performance in every team.
Editorial take: The ability to stay quiet is a meaningful agent feature. A Slack assistant that recognizes the relationship between several messages can find useful work, but it can also become disruptive if it guesses badly. Teams should evaluate missed opportunities and unnecessary interventions together rather than reward an agent simply for posting more often.